What four years of Syracuse service requests say about how a city should measure itself
Syracuse residents filed 60,208 service requests through the SYRCityline app between 2021 and early 2025. We pulled all of them, 21 variables deep, to find what actually drives request volume and where the city could respond better.
he data needed work before it could answer anything. We built a cleaning pipeline that standardized placeholder values, dropped columns more than 70% empty while protecting the timestamp and geographic fields, and imputed missing numeric values at the median. Every imputed value got a flag column, so anyone reading the analysis downstream can see which numbers were observed and which were filled in. We also found a rating of 6 in a 1-to-5 scale, which is the kind of thing that quietly breaks an average if nobody checks.
Two agencies carry two-thirds of the load. Garbage, Recycling and Graffiti accounts for 43.7% of requests and Water and Sewage for 24.3%. Trash and pipes, 68% of everything.
The finding that surprised us. Requests that got a good rating took about 5.7 days longer to close than requests that got a bad one. That does not mean slow service is good. It more likely means the harder problems take longer, and people are satisfied when they are finally solved. The model only explains 4.4% of what drives closure time, so most of the story is something we did not measure. But the direction matters: speed is not a proxy for satisfaction, and a city that manages to close-time is managing the wrong number.
What we would tell the city. Put resources behind the 13208 ZIP code, which generated 5,562 requests, more than any other. Fix the field notes, because over 5,600 descriptions say only “unknown,” which makes those requests untraceable. And stop treating closure speed as the headline metric, because our data says it barely moves how people feel about the outcome.
IST 387, Applied Data Science. Two-person project with Adimchinobi Amaihe. R, dplyr, ggplot2, arules, quanteda.